A task board the agent works from
- Tasks, subtasks and dependencies, with statuses from backlog through review, merging and done. A task that waits on others moves itself along when they finish.
- Every task has a Markdown spec — the thing the agent is actually handed — plus attachments, the questions asked about it, and a summary of what the run did.
- Search across titles, specs and decisions, including semantic search, and a knowledge base the agent reads before it starts and writes to when it learns something.



